Dmitry Babokin

Results 37 issues of Dmitry Babokin

ISPC code may be affected by FTZ/DAZ flags not being set even if program doesn't deal with denormals explicitly, if they appear in the inactive lane as a garbage. The...

Features

ISPC supports 4 math libraries: `default`, `fast`, `svml`, `system`. To make a conscious decision between them user needs information about their precision on specific targets. The standard way to quantify...

Standard Library

ICX/ADL/SPR are not yet supported as separate targets (though they are available as targets for micro-architectural tuning via `--device` switch), but these are chips with different ISAs and it's useful...

Features

```bash > cmake .. -DISPC_PREPARE_PACKAGE=ON -DISPC_INCLUDE_BENCHMARKS=ON -DBENCHMARKS_ISPC_TARGETS=avx1-i32x8 > make -j32 > sde64 -snb -- make test TID 0 SDE-ERROR: Executed instruction not valid for specified chip (SANDYBRIDGE): 0x7fff69f0a309: vpbroadcastb ymm0,...

ISPC language (https://ispc.github.io/) is used in a number of rendering and game dev projects. Some of the projects are fairly large and would benefit from `ccache` support for ISPC. We...

feature

Would be nice to have an ability to explicitly specify dependency for SYCL file via DEPENDS parameter of add_sycl_to_target() in CMake builds.

help wanted
cmake
good first issue

Update for Xe targets including stability, performance, and usability improvements. Performance: - Significant perf boost caused by updates in ISPC optimization pipeline and usage of `newspillcost` finalizer function which dramatically...

Current definition of implicit type conversion gives priority to int64 over float type. I.e. int64 + float is an int64. It's explained in more details in http://ispc.github.io/ispc.html#basic-types-and-type-qualifiers The types rank...

Bugs

```cpp uniform bool foo(uniform bool b) { uniform bool r = b; // r = r + true; // illegal and doesn't work r += true; // illegal, but works....

YARPGen

llvm.ldexp and llvm.experimental.constrained.ldexp intrinsics were introduced in LLVM (current trunk, i.e. 17.0). [eece6ba](https://reviews.llvm.org/rGeece6ba283bd). Need to evaluate them and use for our `ldexp` computation.

Performance
Features